JPWO2005069188A1 - System for predicting interaction between compound and protein, system for predicting similar protein or compound, and method thereof - Google Patents

Abstract

For the interaction of multiple compounds and multiple proteins, construct a database that integrates information on chemical substances and biotechnology, and use them to create a comprehensive interaction analysis method that combines computational speed and accuracy. Established. It interacts with proteins that are arbitrarily assigned and related protein groups based on data that correlates protein amino acid sequence information, compound structure information, and protein-compound interaction information. Identify the compound group that interacts with the protein from the structure-activity relationship model that can distinguish the compound group from any compound group and the compound group that interacts with the protein and its related proteins. A method of predicting by combining with the obtained structure-activity relationship model.

Description

Cross-reference of related applications

  The entire disclosure of Japanese Patent Application No. 2003-435659 (filed on Dec. 26, 2003), including the specification, claims, drawings and abstract, is incorporated into this application by reference to these disclosures. .

  The present invention relates to a system for analyzing and / or predicting an interaction between a compound and a protein, a system for predicting a similar protein or similar compound, and methods thereof. Specifically, based on data in which protein amino acid sequence information, compound structure information, etc., and compound-protein interaction information are correlated with each other, any compound-protein interaction, similar protein or This is a method for analyzing and / or predicting similar compounds.

  In the field of chemical drug discovery, a high-throughput screening system (HTS) that analyzes the interaction of a large number of compounds targeting a specific protein, the synthesis of multiple compounds at the same time, prior to the genome era. As a result of the establishment of combinatorial chemistry, information science and technology (chemoinformatics) related to the biological action of chemical substances that handle a large amount of information has progressed and has achieved some success. On the other hand, in the field of drug discovery in biology, microarray analysis and proteome technology for simultaneously measuring a large number of in vivo factors have been developed. In addition, a reverse proteomics technique for analyzing the interaction of a large number of in vivo factors targeting a specific drug has been established. Along with these experimental methods, bioinformatics techniques such as comparison between base sequences and amino acid sequences and analysis of interaction networks between in vivo factors have progressed.

  However, chemoinformatics and bioinformatics have developed independently, and it is difficult to say that both are well integrated.

  On the other hand, research based on a new paradigm called “genome drug discovery” as one of the drug discovery processes has been energetically promoted around the world. Needless to say, “genome” means “all genes in the living body”. On the other hand, a “drug” interacts directly and indirectly with a plurality of in vivo factors to develop not only the intended pharmacological action but also side effects and toxicity. Therefore, the “genomic drug discovery” paradigm means “considering the interaction between a drug and all factors in vivo”.

  In drug discovery research, an exhaustive interaction analysis (interaction) between a large number of compounds and all in vivo factors is indispensable in order to search for an optimal compound from a large number of compounds. However, since the number of types of compounds is almost infinite, the number of analysis targets is enormous. Therefore, there is a demand for a method for realizing a comprehensive and accurate analysis and / or prediction of a large number of compounds and all in vivo factors with high accuracy and speed.

  Description of the structure of both compound-protein and docking studies that analyze compound-protein complementarity, currently known as a method for analyzing and / or predicting the interaction between proteins and chemical substances that are in vivo factors Informatics technology that performs analysis based on information science using children as explanatory variables is known. Currently, the docking study technique is the most developed, which is a technique for searching for a model in which an arbitrary compound binds well in the vicinity of the active site of a protein. However, in addition to the premise that the three-dimensional structure coordinates of the protein are known, this method requires a great amount of time because it is necessary to search for an optimal solution. It is not suitable for exhaustive analysis from the viewpoint of calculation speed, and its accuracy is not sufficient.

  In recent years, a method for estimating an interaction between an arbitrary compound and a protein without using the three-dimensional structure information has been reported for the problem of the docking study. In other words, it is an informatics analysis in which the structural formula of the compound and the sequence information of the protein are converted into structure descriptors and both are used as input variables.

  The above analysis method does not require a three-dimensional structure, so that the range of application is widened, and an optimal solution search that requires a large amount of calculation time can be avoided, but it has not yet reached a practical level from the viewpoint of calculation accuracy.

  As a method for improving the calculation accuracy, an annotation method for four major classes of ligands and a technique for applying it to in silico screening and library design have been proposed (Non-patent Document 1). The annotation method of this document is based on the hierarchy of ligand functions and existing classifications. Then, a search for the ligand database is executed based on the annotation.

  Further, as a method developed from this, there has been proposed a method for searching for the binding of a ligand to a similar target as well as the same target as the reference ligand (Non-patent Document 2). The search method of this document uses a description of a molecule reflecting the ability of the molecule to interact with the target protein, the molecular structure, and the like.

Ansgar Schuffenhauer et al., “Ontology for Pharmaceutical Ligands and Its Application for in Silico Screening and Library Design,” J. et al. Chem. Inf. Comput. Sci. 2002, 42, 947-955.

Ansgar Schuffenhauer et al., “Similarity Metrics for Ligands Reflecting the Similarity of the Target Proteins”. Chem. Inf. Comput. Sci. 2003, 43, 391-405.

  Comprehensive interaction analysis between in vivo factors (interactome) is positioned as a powerful tool for the purpose of extracting useful information from a huge amount of objects using technologies such as microarrays, proteomes, and bioinformatics. . However, each technique such as bioinformatics including microarray and proteome analysis described above contains many false positives in the obtained information, and it is important to extract truly useful information from them. It has become.

  The problems of the present invention are roughly divided into two problems. One is the construction of a database and the other is the establishment of an informatics method. Specifically, a database that integrates information on chemical substances and biotechnology with respect to the interaction of multiple compounds and multiple proteins, and comprehensive compounds that have both calculation speed and accuracy using them -Establishing methods for analyzing protein interactions.

  The present invention has been developed by combining the two advantages of chemoinformatics and bioinformatics. As a characteristic of the apparatus and method of the present invention, as a result of intensive studies, the present inventors have correlated protein amino acid sequence information, compound structure information, and protein-compound interaction information. Based on the data, (i) a structure-activity relationship model that makes it possible to distinguish a compound group that interacts with a protein group including a protein and its related protein from any other compound group, and (ii) Established a prediction method in combination with a structure-activity relationship model that makes it possible to identify a compound (or compound group) that interacts with a specific protein (or protein group) from a group of compounds that interact with the protein group did.

  The present invention includes a plurality of features as follows.

(1) In the method according to the present invention, protein amino acid sequence information, protein amino acid sequence information systematically classified by function and / or structure similarity, compound structure information, and protein-compound interaction information This is a method for predicting the interaction between an arbitrary protein and a compound based on correlated data.

(2) A method for predicting an interaction between an arbitrary protein and an arbitrary compound according to the present invention includes (a) an interaction with the protein and a protein group that is functionally and / or structurally similar to the protein. A structure-activity relationship model capable of discriminating a group of compounds with respect to an arbitrary group of compounds, and (b) a group of compounds that interact with the protein and a group of proteins functionally and / or structurally similar to the protein Among them, and a structure-activity relationship model that can identify a group of compounds that interact with the protein.

(3) The method for predicting the interaction between an arbitrary protein and an arbitrary compound according to the present invention is based on systematic classification based on similarity in function and / or structure of amino acid sequence information. A structure-activity relationship model capable of discriminating a compound group that interacts with a protein group belonging to the highest class item to which it belongs, with respect to any compound group, and (b) a lower level than the highest class item to which the protein belongs A compound group that interacts with a protein group that belongs to a class item to which the protein belongs, and a compound that interacts with a protein group that belongs to a child class item that has a common parent class item with the protein This is a method of predicting by combining with a structure activity model that can be identified for a group.

(8) A system according to the present invention is a prediction system for predicting a protein having a similar function and / or structure, and (a) a first class belonging to the first class indicating the classification of the functional characteristic and / or structural characteristic of the protein. A first recording means for recording information on one class protein and information on a non-first class protein not belonging to the first classification; (b) a selection from among the first classification proteins recorded on the first recording means Information of the second class protein belonging to the second class indicating the classification of the functional feature and / or the structural feature which is a sub-concept than the first class, and belonging to the second class but belonging to the second class Second recording means for recording non-secondary classification proteins, (c) acquisition means for acquiring prediction target information indicating functional characteristics and / or structural characteristics of the prediction target protein, (d) prediction acquired by the acquisition means Target information and First analysis means for analyzing similarity between the protein to be predicted and the first classification protein in comparison with the non-first classification protein based on the information recorded in the first recording means, (e ) When it is analyzed by the first analysis means that the protein to be predicted is similar to the first class protein, the prediction target information and the information recorded in the second recording means are further used. Second analysis means for analyzing the similarity between the protein to be predicted and the second classification protein in comparison with the non-second classification protein, (f) based on the analysis results by the first analysis means and the second analysis means Output means for outputting information on a protein having functional characteristics and / or structural characteristics similar to the protein to be predicted,
It is the prediction system provided with.

(9) In the prediction system of the present invention, (a) protein information recorded in the first recording means and / or (b) protein information recorded in the second recording means interacts with the protein. Associated with compound information,
The prediction system is further similar to (g) information on a compound that interacts with the protein, and (d) the protein to be predicted analyzed by the first analysis means and / or (e) the second analysis means. Interaction information analysis means for analyzing information on a compound predicted to interact with the protein to be predicted based on the protein information, and (f) the output means has the function of the protein to be predicted and the function Information on proteins having similar characteristics and / or structural features and / or information on compounds analyzed by the interaction information analysis means are output.

(10) The (f) output means of the present invention is more conceptual than the second classification in addition to the information of the protein and / or the compound that interacts with the protein as an analysis result by the second analysis means. The information on the protein and / or the compound that interacts with the protein as an analysis result by the first analysis means is also output.

(13) The system according to the present invention is a prediction system for predicting a protein (or compound) having a similar function and / or structure, and (a) a classification of the functional characteristic and / or structural characteristic of the protein (or compound). First information that records information on a first class protein (or first class compound) belonging to the first class and information on a non-first class protein (or non-first class compound) not belonging to the first class Recording means, (b) a functional and / or structural feature that is selected from the first class proteins (or first class compounds) recorded in the first recording means and is a sub-concept than the first class Information of the second class protein (or second class compound) belonging to the second class indicating the classification of the non-second class protein (or non-second class compound belonging to the first class but not belonging to the second class) ) (C) acquisition means for acquiring prediction target information indicating the functional characteristics and / or structure characteristics of the prediction target protein (or prediction target compound), (d) the prediction target acquired by the acquisition means Based on the information and the information recorded in the first recording means, the prediction target protein (or prediction target compound) and the first in comparison with the non-first classification protein (or non-first classification compound) First analysis means for analyzing similarity to a classification protein (or first classification compound); (e) the prediction target protein (or prediction target compound) is converted into the first classification protein (or first classification) by the first analysis means; If it is analyzed that it is similar to the classification compound), the non-second classification protein (or non-second classification) is further based on the prediction target information and the information recorded in the second recording means. Second analysis means for analyzing the similarity between the prediction target protein (or prediction target compound) and the second classification protein (or second classification compound) in comparison with a similar compound), (f) the first analysis An output means for outputting information on a protein (or compound) having a functional characteristic and / or a structural characteristic similar to that of the prediction target protein (or prediction target compound) based on the analysis result of the means and the second analysis means;
It is the prediction system provided with.

  Features, other objects, applications, effects, etc. of the present invention will become more apparent from the following disclosure in view of the drawings.

FIG. 1 is a flowchart showing a processing procedure according to an embodiment of the present invention. FIG. 2 is a diagram showing a concept of a category hierarchy when proteins in the embodiment are classified hierarchically. FIG. 3 is an example of a configuration of a screen displaying the analysis result according to the embodiment. FIG. 4 shows a global model (Global Model) of phosphodiesterase (hereinafter referred to as “PDE”) + subtypes among the models obtained for the active group data set in Example 1. It is the graph which showed the difference in discriminability at the time of applying "Local model of (Local Model)" and "Global model of each subtype". The vertical axis of the graph indicates “rate recognized as activity”. FIG. 5 shows “PDE global model + local model of each subtype” and “global of each subtype” among the models obtained for the inactive group data set used in the global model in Example 1. It is the graph which showed the difference in discriminating ability at the time of applying a "model". The vertical axis of the graph indicates “rate recognized as activity”. FIG. 6 shows “PDE global model + local model of each subtype” and “global model of each subtype” among the models obtained for the inactive group data set used for the local model in Example 1. It is the graph which showed the difference in discriminating ability at the time of applying "." The vertical axis of the graph indicates “rate recognized as activity”. FIG. 7 is a graph showing the results of comparing the discrimination ability obtained from the CART method (prior equiprobability) and the Bayesian network analysis for the global model of PDE and the local model of each subtype for the active group data set. It is. The vertical axis of the graph indicates “rate recognized as activity”. FIG. 8 shows the results of comparing the discrimination ability obtained from the CART method (prior probability) and the Bayesian network analysis for the global model of PDE and the local model of each subtype for the inactive group data set. It is a graph. The vertical axis of the graph indicates “rate recognized as activity”. FIG. 9 is a graph showing the result of comparing the CART method (equal establishment) and Bayesian network analysis for the PDE global model and the local model of each subtype for the active group data set. The vertical axis of the graph indicates “rate recognized as activity”. FIG. 10 is a graph showing the result of comparing the CART method (equal establishment) and Bayesian network analysis for the global model of PDE and the local model of each subtype for the inactive group data set. The vertical axis of the graph indicates “rate recognized as activity”. FIG. 11A is a combination of an active group, an inactive group used for the global model, and an inactive group used for the local model in combination with a global model of PDE and a local model of subtype (PDE1). It is a distribution graph. The vertical axis of the graph indicates “rate recognized as activity”. FIG. 11B is a combination of an active group, an inactive group used for the global model, and an inactive group used for the local model in combination with the global model of PDE and the local model of the subtype (PDE2). It is a distribution graph. FIG. 11C is a combination of a PDE global model and a subtype (PDE3) local model applied to three types of data sets: an active group, an inactive group used for the global model, and an inactive group used for the local model. It is a distribution graph. FIG. 11D is a combination of a global model of PDE and a local model of subtype (PDE4), and applied to three types of data sets: an active group, an inactive group used for the global model, and an inactive group used for the local model. It is a distribution graph. FIG. 11E is a combination of a PDE global model and a subtype (PDE5) local model applied to three types of data sets: an active group, an inactive group used for the global model, and an inactive group used for the local model. It is a distribution graph. FIG. 12 is a functional block diagram of the interaction analysis device. FIG. 13 is a hardware configuration example of the analysis apparatus. FIG. 14 is a conceptual diagram of an interaction analysis process as an embodiment of the present invention. FIG. 15 is a diagram illustrating a structure example of a protein database according to the embodiment. FIG. 16 is a diagram illustrating a structure example of a compound database in the embodiment. FIG. 17 is a diagram illustrating a structure example of an interaction database according to the embodiment. FIG. 18 is a diagram illustrating a structure example of a systematic classification database in the embodiment. FIG. 19 is a schematic diagram illustrating an evaluation function in the embodiment. FIG. 20 is a flowchart of the interaction analysis processing program according to the first embodiment. FIG. 21 is a flowchart of the interaction analysis processing program according to the second embodiment. 22A and 22B are screen display examples output by the second embodiment.

  The “protein amino acid information” in the present invention includes, for example, a sequence, a function, or a three-dimensional structure. Examples of sequences and functions include known information, information estimated from bioinformatics, multiple types of annotation information, ontology information directed to systematic function classification, and the like. As for the three-dimensional structure, known information includes a PDB (Protein Data Bank) as a public database, a commercial or in-house database constructed by homology modeling, and the like. Commercial homology modeling databases include FAMSBASE sold by SGI.

  “Structural information of a compound” of the present invention includes, for example, information describing a structural formula, drug information, and the like. For example, presence or absence and / or strength of pharmacological activity on a compound, or development from Biological Testing to Launched Includes stage information. For example, MDDR (MDL Drug Data Report) of MDL is mentioned.

  The “protein amino acid information” used in the present invention is a database obtained by combining and combining the amino acid information parts of the protein or a database including all the amino acid information of the protein and classified. Any source may be used as long as it is commercial or in-house. Preferably, information systematically classified by function and / or structural similarity is mentioned, for example, Gene Ontology (registered trademark) information and the like. “Gene Ontology” is published on the Gene Ontology Consortium website (http://www.geneontology.org).

  The “interaction” of the present invention indicates a relationship between a protein and a compound that exhibits activity against the protein, or a relationship between a compound and a protein that is complementary to the compound. “Interaction” includes the interaction of a plurality of compounds with a plurality of proteins or the interaction of a plurality of proteins with a plurality of compounds.

  The “structure-activity relationship model” of the present invention uses, for example, a structural feature information of a protein (or compound) and an arbitrary protein (or compound) belonging to a predetermined classification as a data set, gives explanatory variables to them, Using the evaluation function (hereinafter referred to as the Global Model) obtained by the analysis method of, and the structure feature information of the protein (or compound) belonging to the predetermined classification and the related protein (or compound) as the data set, And an evaluation function (hereinafter referred to as a local model (local model)) obtained by a predetermined analysis method. As the explanatory variable, for example, a pharmacophore descriptor used as an explanatory variable in quantitative structure-activity relationship analysis, a topological index used for similarity search, an index related to ADMET, or the like can be used. As an analysis method, for example, multiple regression analysis, linear / nonlinear discriminant analysis, logistic regression analysis, neural network, decision tree analysis, Bayesian network, or support vector machine can be used.

  The functions of the apparatus according to the embodiment of the present invention can be divided into “search”, “browse”, and “analysis”, and the existing environment can be used as it is for “search” and “analysis”. As an existing environment, a system that has an input / output function for a compound and protein list and can be browsed in a format in which compound information and bioinformation are associated with each other is preferable. Examples thereof include a client server type system and a web-based system. The system of the present invention is capable of accessing a plurality of databases and individually specifying a display reflecting an input list and an output target. Examples of the program description language include C, C ++, JAVA (registered trademark), HTML, and XML. Existing programs such as “Chime” provided free of charge by MDL can also be used to browse structural formulas on the Web base.

  Hereinafter, embodiments of the present invention will be described.

Table of contents Outline of Embodiments 2. 2. Verification of analysis processing method 3. Outline of interaction analysis apparatus and analysis method Database 5. Interaction analysis processing Other embodiments -------------------
1. 1. Outline of Embodiment 1-1. Database and Interaction Analysis Processing FIG. 1 is a flowchart showing the concept of one embodiment of the present invention. Reference numeral 101 denotes a database of information on the protein side. The information on the protein side to be integrated includes “amino acid sequence” and “three-dimensional structure (including modeled one)”, for example, information obtained from SwissProt or the like. Reference numeral 102 represents a database of information on the compound side. The information on the compound side to be integrated includes “structural formula” and “conformation”, for example, information obtained from CAS or the like. Reference numeral 103 denotes amino acid sequence information of proteins systematically classified by function and / or structural similarity. For example, there is a category when proteins are classified hierarchically, such as ontology information including the GO number of Gene Ontology. The information of the protein 101 is related to the systematic classification information by the information 103. Reference numeral 104 denotes an interaction database associated with information on proteins and compounds. Examples of the information include commercial databases such as MDDR (MDL Drug Data Report) of MDL, pharmacological activity test data, reverse proteomics, and the like. Reference numeral 105 denotes a structure-activity relationship analysis function.

  In the embodiment, an integrated database (101 and 103 in FIG. 1), structure information (102 in FIG. 1), and interaction information between a protein and a compound are associated with each other. Based on the data of 104) in FIG. 1, a comprehensive interaction analysis is performed in consideration of both the commonality and difference of the functional features and / or structural features to be analyzed. Specifically, using amino acid information of proteins systematically classified by 103 functions and / or structural similarity, for example, between child nodes having a common parent node in Gene Ontology (registered trademark) Build a selective structure-activity relationship (SAR) model at all levels. Then, the suitability (for example, the presence or absence of interaction) of the analysis target with respect to those models is evaluated. The node indicates a category when the proteins are hierarchically classified, and examples thereof include a gene ontology GO number. FIG. 2 shows an image of the node hierarchy. The analysis model in each node is a combination (203) of a “global model in a node one level higher” (201) and a “local model between peer nodes” (202). The global model (204) and the local model (205) of each node ”. Model construction at each node is performed by informatics analysis using various structure descriptors as explanatory variables. As the various structure descriptors, a pharmacophore descriptor used as an explanatory variable in quantitative structure-activity relationship analysis, a topological index used for similarity search, an ADMET-related index, or the like can be used. The “global model of the highest node” in FIG. 1 represents a model that can significantly distinguish the compound group belonging to the highest node from any other compound group. The “local model of each node” represents a model that can preferentially identify a compound group belonging to a certain node with respect to a compound group belonging to another node having a common parent node.

1-2. Interface FIG. 3 shows a screen display image of the interaction analysis system according to the embodiment. The contents of the interaction analysis process will be described later. 301 in FIG. 3 shows a dendrogram of functional classification of amino acid sequence information (eg, gene ontology) of proteins systematically classified by functional and / or structural similarity. 301 is associated with the corresponding compound number. In the default display, the tree is expanded only for nodes including the designated amino acid sequence information number (for example, the GO number of Gene Ontology) or the compound number, and the rest are displayed in a folded state. The tree on the right side of FIG. 3 shows the expanded state. For example, the GO number corresponding to the analysis result is represented by a different character color. In each node, the lower amino acid sequence information number (for example, “GO number”, the same applies hereinafter), the number of amino acids, and the total number of compounds are displayed, and are changed as needed according to the change in the display format. Clicking on any compound number will display the corresponding structural formula and associated data. Four buttons having a list input / output function are arranged (302 to 305 in FIG. 3). The input assumes amino acid sequence information numbers and compound numbers, and the output assumes protein sequence information, protein coordinate data (PDB format), and compound numbers. A check box is set for each of the amino acid sequence information number and the compound number corresponding to the terminal node, and a list of the checked items is output.

  Next, the function of each display button will be described using the reference numerals in FIG. A plurality of amino acid sequence information numbers 301 can be designated by clicking a node or inputting a number list of amino acid sequence information. Multiple structural formula queries (Query) 303 can be specified, and the number of the specified amino acid sequence information x the structural formula score is calculated by entering in the editor or pressing the Run button 304 for specifying the SD file. Is done. When the number of the specific amino acid sequence information of 301 or one of the compounds displayed in 306 is designated, the score is displayed on the other. When the filter button (305) is pressed after specifying the threshold value, records (nodes) that are equal to or higher than the threshold value are extracted. The filter operation 305 can be executed a plurality of times with designation of “and / or / not”, and the result can be output to a delimited text file such as CSV (Comma Separated Values) format.

2. Verification of Analysis Processing Method Verification of the analysis processing method used by the interaction analysis system as the embodiment will be described. Hereinafter, verification results of a plurality of analysis processing methods using examples of mutual analysis information of predetermined compounds and proteins will be shown.

2-1. Analysis by phosphodiesterase (Phosphodiesterase) by CART method The interaction analysis processing by the interaction analysis system as an embodiment is a CART method using pharmacophore descriptors for identifying the presence or absence of pharmacological activity from structural formula information of compounds as explanatory variables. Is used. In the embodiment, as an example, a “global model (global model)” using various compound sets including an “inactive group” as a compound group that does not interact with the target protein (or “small interaction”. The same applies hereinafter). "And a" local model (local model) "using a compound set near the active group, the interaction analysis process is realized. The contents of the global model and the local model will be described later.

  Table 1 shows the discriminating ability of each lower node between the “global model of each node” and the “global model of each node and the local model of each node” with respect to the one-level tree structure that is the basic unit of the gene ontology. A comparison is shown. In the experimental example, five subtypes (PDE1 to PDE5) having a phosphodiesterase (Phosphodiesterase (hereinafter referred to as “PDE”)) as an upper node are used as an example. PDE is a general term for enzymes that hydrolyze phosphodiester to phosphomonoester.

  The 2871 compound is assigned to the PDE which is the upper node. The number of compounds belonging to each lower node varies widely, with a minimum of 29 compounds (PDE2) and a maximum of 1699 compounds (PDE4). Only compounds with a molecular weight of 200 or more and less than 800 were adopted in the inactive group included in the “global model” so that there would be no distribution difference due to molecular size or the like. The inactive group of the “global model” used 3000 compounds randomly extracted from about 500,000 commercially available HTS compounds collected from many vendors. In the creation of the “local model”, it was regarded as inactive except for those with known pharmacological activity among the 2871 compounds belonging to the upper node. In the comparison results illustrated in Table 1, all the analyzes use common parameters (maximum hierarchy = 10 / parent node = 5 / child node = 1), and only the prior probability is “data set dependent” (“cart_data”). Both “equal probabilities” (“cart_even”) were examined. The analysis results by the CART method of PDE (upper node) and each subtype (lower node) of PDE are shown in Table 1 and FIGS.

  The upper part of Table 1 shows the discrimination ability of the global model, and the lower part shows the discrimination ability of the local model. Two numbers are written in each column, but the left side is the discriminating ability for the data set (learning data) used to build the model (e.g., evaluation function for identifying compounds), the right side is The discriminating ability for the data set (verification data) used for model verification is shown. An overview of the results shows that generally favorable models are obtained when the prior probabilities are made equal. For the following examination, a model with prior probabilities made equiangular was used.

  4 to 6 are: (1) Compound group interacting with PDE (active group), (2) Inactive group used in global model (compound group not interacting with PDE), (3) Used for local model For the three types of data set of the inactive group, “(a) PDE global model + local model of each subtype (displayed as“ Global_PDE & Local _PDEx ”in the figure) ) ”And“ (b) Global model of each subtype (displayed as “Global_PDEx” in the figure) ”are graphs showing the difference in discriminability. As shown in FIG. 4 and FIG. 5, for the data sets of (1) active group and (2) inactive group used in the global model, (a) and (b) there is little difference in discriminability between both. Was not seen. However, as shown in FIG. 6, (3) the inactive group data set used in the local model shows very poor discrimination ability when using “(b) global model of each subtype”. There wasn't. This means that the global model cannot analyze differences between related proteins. From these considerations, in order to determine whether or not an arbitrary compound interacts with a protein belonging to each node, a “global model at a node one level higher” and a “local model between lower nodes” are determined. The necessity of combining was suggested.

2-2. Analysis by Bayesian network When evaluating an arbitrary compound, the evaluation is performed in stages from the upper node to the lower node. However, in the method of determining the classification in binary as in the CART method, a compound that has been falsely determined by the upper node is not evaluated by a node below that. In order to deal with this point, for example, a technique of determining as a score value can be adopted instead of a technique of determining classification in binary.

  Analysis similar to the above “2-1” was performed by Bayesian network analysis (Belief Network) in which specific objective variables were set. First, the PDE global model and local models of each subtype were compared with the discriminating ability obtained from the CART method (prior equiprobability). In general analysis software BayesiaLab 2.0, since five kinds of analysis methods can be used, these comparative studies were also performed (see Table 2).

  As shown in the table, the Naive Bayes method, Markov Blanket method, and Augmented Markov Blanket method, which complete the calculation in a short time, showed similar trends, but their discriminating ability was not sufficient. . The Sons & Spouses method, on the other hand, requires a relatively long calculation time compared to the previous three methods, but shows discrimination ability similar to that of the CART method. However, when the number of active groups is extremely small, the discrimination ability is greatly reduced. In contrast, the Augmented Naive Bayes method required almost the same calculation time, but showed high discrimination ability even when the number of active groups was small. On the other hand, for the global model with a large number of samples, the Sons & Spouses method showed the same discrimination ability as the CART method, but it was clearly overlearned. Therefore, Augmented Naive Bayes method and Sons & Spouses method can be said to have advantages and disadvantages.

  Here, in the CART method described in the section “2-1.”, Improvement of discrimination ability was seen by setting prior probabilities to equal probabilities. This was noticeable when the amount of data in the active group was very small compared to the inactive group. Therefore, we examined Bayesian network analysis with equal prior probabilities. Since BayesiaLab does not have a function to set prior probabilities, the results output as probabilities are considered externally. More specifically, if the objective variable is 2 classes, it is determined which class the probability variable belongs to at a boundary of 0.5. If the prior probabilities of the two classes are 1/10 and 9/10, an effect equivalent to the prior equiprobability in the CART method can be expected by setting the boundary value of the class to 0.1. Table 3 shows the results of Bayesian network analysis.

  Naturally, the discrimination rate for the active group was improved, and the misrecognition rate for the inactive group was increased. In particular, PDE-1 and PDE-2, which have a small number of data, showed a significant improvement in recognition rate. The Augmented Naive Bayes method was over-learning, so there was no difference in the results even when considering the prior probabilities. These results suggest that good classification results can be obtained by determining the classification threshold for each model. For example, the data set is divided into learning and testing, a model (for example, an evaluation function for identifying a compound) is constructed from the learning data, and the score value obtained by evaluating the test data using the model is calculated. The classification threshold may be determined from the distribution. As an example of the analysis method, the Sons & Spouses method can be adopted in consideration of the balance between discrimination ability and over-learning. The results are shown in FIGS.

  Next, the PDE global model obtained in the above study and the local model of each subtype are combined to (1) active group, (2) inactive group used in the global model, and (3) local model. It was applied to three different data sets of the inactive group. In the CART method, classification is determined by binary, but here, it is expressed as a probability of matching at each node. The probability is expressed as a conditional probability with the upper node as an example, and the classification threshold is determined from the distribution of probability values. Each of FIG. 11A, B, C, D, and E shows the probability distribution of three types of data sets in each subtype (PDE-1 to PDE-5). In the figure, “1” represents an active group, “0” represents an inactive group used in the local model, and “−1” represents an inactive group used in the global model. The horizontal axis represents the conditional probability when applying the global model of PDE and the local model of each subtype. The vertical axis of the graph indicates “rate recognized as activity”. As shown in FIG. 11, both the inactive group of the global model and the local model are well separated from the active group, similarly to the result of the CART method. Moreover, since it is represented by a conditional probability, there are some which take an intermediate value of 0-1.

2-3. Analysis by support vector machine Next, we examined the handling of inert information when building a local model. Normally, only those with known pharmacological activity are stored in the database. Therefore, it is impossible to know whether a substance without pharmacological activity information is actually inactive or has not been investigated. Regarding the construction of the global model, in the embodiment, various compound sets as inactive information are given as a data set, so that there is no problem in terms of probability. On the other hand, regarding the construction of a local model, in the embodiment, related compounds are given as a data set. Therefore, there is a non-negligible percentage of missing data as it is not tested despite having pharmacological activity. In order to avoid this problem, for example, an inactive information may be treated as a missing value, and an analysis model using only data having a known pharmacological activity as a data set may be constructed. Therefore, an inhibitor activity model for PDE and subtypes of PDE is a Support Vector Machine. Hereinafter, it is referred to as “SVM”. ) Was used to construct a prediction model and cross-validation (4-fold cross validation) was performed. The parameters in SVM were fixed, standardization of explanatory variables and Gaussian kernel were used. The software used is LIBSVM. The concept of SVM is described in, for example, “Vapnik, Statistical Learning Theory, Wiley, 1998”. Tables 4 and 5 show the results of cross-validation.

  OCSVM (One-Class SVM) is a learning algorithm that identifies the activity / inactivity of a drug using only the feature quantity (descriptor) of the active group. An OCSVM model was constructed for inhibitors of PDE1-5 to verify cross-validation within the active group and the ability to discriminate against 3000 randomly sampled compounds. The parameters in SVM were fixed, and the standardization of explanatory variables and the RBF kernel (Gaussian kernel) were used. The software used is LIBSVM. The concept of OCSVM is described in, for example, “B. Scholkopf, et.al. Estimating the support of a high-dimensional distribution. Neural Computation, 13, 2001, 1443-1471”.

  The computer experiment was performed according to the following procedure.

(1) Verification of discrimination ability of each PDE subtype by cross-validation method by OCSVM (2) Verification of discrimination ability of each PDE subtype by the same data as training data (3) Verification of discrimination ability for randomly sampled compounds The results of the experiment are shown in Table 6.

  Heretofore, the results of verification of the analysis processing method have been described using a plurality of general statistical processes as examples. The analysis processing according to the present invention can be realized by any one of the above-described methods, a modification of each method, a combination of the methods, or a method known to those skilled in the art. In the following description, an apparatus for realizing the above-described analysis processing method and details of the analysis processing method will be mainly described as an embodiment of the present invention.

3. 3. Outline of interaction analysis apparatus and analysis method 3-1. Functional Block FIG. 12 shows a functional block diagram of an interaction analysis apparatus 500 as an embodiment of the system or method of the present invention. The interaction analysis apparatus 500 includes (a) first recording means 72, (b) second recording means 74, (c) acquisition means 70, (d) first analysis means 76, (e) second analysis means 78, (F) An output unit 82 and (g) an interaction information analysis unit 80 are provided.

3-2. Hardware Configuration FIG. 13 shows an example of a hardware configuration in which the interaction analysis apparatus 500 shown in FIG. 12 is realized using a CPU. The interaction analysis device 500 includes a CPU 10, a memory 12, a speaker 14, a communication circuit 16, a keyboard / mouse 18, a display (display device) 20, and a hard disk 22.

  The CPU 10 executes an interaction analysis process described later, and controls the entire interaction analysis apparatus 500. The hard disk 22 records a protein database 600, a compound database 700, an interaction database 800, a systematic classification database 900, and a program for controlling the interaction analysis device 500 (for example, an interaction analysis processing program). The memory 12 is used as a work area for the CPU 10 and a storage area for acquired data. Information input by operating the keyboard / mouse 18 is processed by the CPU 10.

  In the embodiment, as an example of an operating system (OS) of the interaction analysis apparatus 500, Windows (registered trademark) XP, NT, 2000, or the like of Microsoft Corporation is used. The computer program of the embodiment realizes each function shown in FIG. 12 in cooperation with the OS. However, the present invention is not limited to this, and each function may be realized by a computer program alone.

3-3. Analysis Method FIG. 14 is a conceptual diagram of an interaction analysis process as an embodiment of the present invention. The interaction analysis apparatus 500 as an embodiment includes a protein database 600, a compound database 700, an interaction database 800, and a systematic classification database 900. For example, the apparatus 500 has a function of predicting a protein that interacts with a compound to be analyzed and a function of predicting a compound that interacts with a protein to be analyzed.

  Information relating to a plurality of proteins is recorded in the protein database 600. Information relating to a plurality of compounds is recorded in the compound database 700. Information on the interaction between the protein and the compound is recorded in the interaction database (symbol 1000). Accordingly, the proteins recorded in the protein database 600 and the compounds recorded in the compound database 700 are associated with each other.

In the embodiment, the protein information recorded in the protein database 600 is systematically classified by the information of the systematic classification database 900. As another embodiment, the systematic classification database 900 may systematically classify compound information recorded in the compound database 700. Alternatively, the systematic classification database 900 may systematically classify information that combines a protein (included in the database 600) and a compound that interacts with the protein (included in the database 700). The systematic classification database 900 according to the embodiment includes, as an example, information obtained by systematically classifying information related to a protein based on similarity in function and / or structure of the protein, more specifically, a gene ontology GO number. It includes information that classifies proteins hierarchically by ontology information that includes etc. As described above, the interacting database 800 associates the interacting protein and compound, so that the interaction information between the protein and the compound is systematically classified based on the information in the database 900 ( Symbol 1002). The systematic classification of proteins and / or compounds is not limited to those described in the embodiments, and includes, for example, physical properties, molecular structures, structural formulas, amino acid sequences, structural annotation information, ligand functions, or functional annotation information. Similarity of information about the structure can be used.

  A tree structure 1004 shown in FIG. 14 shows the relationship between proteins and / or compounds that are systematically classified by the systematic classification database 900. The upper classification node 1008 includes a plurality of proteins and / or compounds. On the other hand, each of the lower classification nodes 1006 and 1010 includes those having predetermined functional characteristics and / or structural characteristics selected from proteins and / or compounds belonging to the higher classification node 1008. FIG. 14 shows a total of three classification nodes divided into two layers for convenience of explanation. Arbitrary numbers can be adopted as the number of layers in the systematic classification and the number of classification nodes included in each layer according to the contents of the systematic classification to be used.

  When analyzing a compound and / or protein that interacts with a protein and / or compound to be analyzed, the interaction analysis apparatus 500 obtains information on each node of the protein and / or compound systematically classified by the tree structure 1004. Use. Specifically, the device 500 analyzes whether or not the analysis target belongs to the higher class node (step S101). Next, the apparatus 500 analyzes whether or not the analysis target belongs to the lower classification node (S103). As described above, the apparatus 500 analyzes the presence / absence of attribution to each classification node with respect to the analysis target, that is, a protein and / or compound having similar functions and / or structures (interaction information included in the database is known). Information on the protein and / or compound that interacts with the analysis target.

3-4. Explanation of Device Functions As a correspondence between a part of the functions of each configuration of the interaction analysis apparatus 500 shown in FIG. 12 and each function in the embodiment, for example, the following contents can be given.

  The 1st recording means 72 respond | corresponds to the information regarding the node A recorded on the systematic classification | category database 900 (refer FIG. 18) (refer the table 66 of FIG. 19). The 2nd recording means 74 respond | corresponds to the information regarding the node A-1 (or A-2) recorded on the systematic classification | category database 900 (refer the table 62 or 68 of FIG. 19). The acquisition unit 70 corresponds to the CPU 10 of the device 500 that executes the process of step S201 in FIG. The first analysis unit 76 corresponds to the CPU 10 that executes the process of step S203 of FIG. The second analysis unit 78 corresponds to the CPU 10 that executes the process of step S205 of FIG. The output unit 82 corresponds to the CPU 10 that executes the process of step S211 of FIG. 20 or step S307 of FIG. The interaction information analysis means 80 corresponds to the CPU 10 that executes the process of step S305 in FIG.

4). Database 4-1. Protein Database The contents recorded in each database recorded on the hard disk 22 of the interaction analysis apparatus 500 will be described. FIG. 15 shows the recorded contents of the protein database 600 as an embodiment. In the protein database 600, information on a plurality of proteins is recorded. Specifically, the protein database 600 includes information on a “protein ID (Protein ID)” for specifying a protein and a “structure index” as an example of the structural characteristics and / or functional characteristics of the protein. Column is included. Information on each protein included in the protein database 600 is based on information in a general public database. The “structure index” is, for example, an amino acid sequence and / or three-dimensional structure information of a protein that is digitized by means well known to those skilled in the art.

4-2. Compound Database FIG. 16 shows the recorded contents of the compound database 700 as an embodiment. In the compound database 700, information on a plurality of compounds is recorded. Specifically, the compound database 700 includes a “compound ID” that identifies a compound, and a column that records information indicating the structural characteristics and / or functional characteristics of the compound. The information indicating the structural characteristics and / or functional characteristics of the compound includes, for example, structural characteristic information obtained by quantifying the structural characteristics based on the function (including physical properties) of the compound and / or the structural formula of the compound. In FIG. 16, LogP (oil-water partition coefficient, n-octanol / water partition coefficient)), hydrogen bond acceptor (HBA), hydrogen bond donor (HBD) are exemplified as structural feature information. )), Molecular weight (MW). Information on each compound included in the compound database 700 is based on information in a general public database.

4-3. Interaction Database FIG. 17 shows recorded contents of an interaction database 800 as an embodiment. In the interaction database 800, a protein (specified by “Protein ID”) included in the protein database 600 and a compound (“Compound ID”) specified in the compound database 700 are specified. "Activity", which is information related to the interaction (for example, information on a compound exhibiting pharmacological activity with respect to a protein). As this activity information, for example, information on a general public database and / or experimentally confirmed information can be used in addition to information on MDL (MDL Drug Data Report) of MDL. In addition, this interaction information can also be created based on the correspondence between each name (including synonyms) of a protein and a compound exhibiting pharmacological activity. In the figure, for example, IDs “P001” and “C005” interact (“Activity = 1”), and “P002” and “C123” do not interact (“Activity = 2”). As another embodiment of the information to be recorded in the “activity” column, a numerical value (including a score value indicating a probability) serving as an index of interaction can be recorded.

  In the embodiment, information related to the interaction between the protein and the compound is recorded in the interaction database 800. In other embodiments, the information on the interaction can be recorded in the protein database 600 and / or the compound database 700 to associate the combination of interacting protein and compound. As another embodiment, the apparatus 500 can also analyze the interaction between proteins. In this case, the protein database 600 and / or the interaction database 800 records combinations of interacting proteins.

4-4. Systematic classification database 900
FIG. 18 shows recorded contents of the systematic classification database 900 as an embodiment. The systematic classification database 900 includes information for systematically classifying a plurality of proteins recorded in the protein database 600 according to function and / or structural similarity. In the embodiment, as an example of systematic classification based on similarity in function and / or structure, proteins are hierarchically classified according to functional classification information (for example, GO number of Gene Ontology) of amino acid sequence information. The systematic classification database 900 records protein systematic classification information based on similarity in function and / or structure in, for example, an XML (Extensible Markup Language) tree structure 50.

  Each node of the XML tree structure 50 is associated with a node number based on the GO number of the gene ontology and an evaluation function. The table data 52 recorded in the systematic classification database 900 records the correspondence between the protein ID and the node number included in the XML node. The table data 54 recorded in the systematic classification database 900 records a correspondence between a node number and an evaluation function for determining attribution to the node. In the example illustrated in FIG. 18, the protein IDs “P001”, “P002”, and “P003” are included in the node number “A-1”. Whether or not an arbitrary protein belongs to the node “A-1” is determined by the evaluation function y = fA−1 (x).

  FIG. 19 is a schematic diagram illustrating an evaluation function in the embodiment. By using the evaluation function (global model and local model) as an embodiment of the “structure-activity relationship model” of the present invention, based on the function and / or structural feature information of the protein (or compound) as the analysis target, Identifying the protein database 600 and / or compound database 700 having functions and / or structural characteristics similar to that of the protein (or compound), and compounds interacting with the protein (or compound) ( Or protein). When the analysis target is a protein, an evaluation function having the function information and / or structural feature information of the protein as explanatory variables is used. On the other hand, when the analysis target is a compound, an evaluation function having the function and / or structure characteristic information of the compound as explanatory variables is used.

  FIG. 19 illustrates an evaluation function using protein structural feature information as an explanatory variable as an example. The tree structure 60 is protein systematic classification information recorded in the systematic classification database 900 shown in FIG. The table 66 includes proteins belonging to the higher-level node A (“P001” to “P006” indicated by the symbol 67) and arbitrary proteins (“P007” to “). On the other hand, the tables 62 and 68 include proteins belonging to the node A (“P001” to “P006”).

  The evaluation function can be obtained by using a predetermined analysis method based on information on a protein (or compound) whose structural feature information is known, which is included in the protein database 600 (or the compound database 700). In the embodiment, the function that the apparatus 500 creates an evaluation function is expressed as a “learning function”. As shown by the symbol 69, the evaluation function of the classification node A can discriminate between proteins belonging to the classification node A and arbitrary proteins not belonging to the classification A when the structural feature information of the protein is given as the explanatory variable X. This function is If the presence / absence of the node is Y, the evaluation function is expressed as Y = fA (X). Symbol 64 indicates a protein belonging to node A-1 (“P001”, “P002”, “P005”) and a protein not belonging to node A-1 (“P003” “ P004 "" P006 ") is shown as an evaluation function (evaluation function: Y = fA-1 (X)). Symbol 65 indicates an evaluation function that makes it possible to significantly distinguish a protein belonging to node A-2, which is a lower node, from a protein belonging to classification node A, and a protein not belonging to node A-2 (evaluation function: Y = fA-2 (X)). Due to the nature of the classification, there are cases where proteins belonging to both nodes A-1 and A-2 exist.

  As described above, the evaluation function indicated by symbol 69 is different from the evaluation function indicated by symbols 64 and 65 in the data set used for obtaining the evaluation function. In the upper node (node A), an evaluation function is obtained using information on proteins belonging to a predetermined upper classification (node A) and information on arbitrary proteins as a data set. On the other hand, in the lower node (node A-1), information on the protein belonging to the lower classification (node A-1) and information on the related protein (belonging to node A but not belonging to node A-1) Get the evaluation function using. In the embodiment, the evaluation function included in the symbol 69 is expressed as a global model, and the evaluation function included in the symbols 64 and 65 is expressed as a local model.

  One feature of the interaction analysis processing described below is that the global model in the upper node and the local model in the lower node are executed in combination. More specifically, the global model and the local model use different data sets. Therefore, it is possible to narrow down the classification nodes of the analysis target in the wide comparison target range by the global model, and specify the classification node of the analysis target after making the difference from the comparison target of the neighborhood comparatively significant by the local model. be able to. In other words, in the present embodiment, analysis (global model) in consideration of “commonality” of functional features and / or structural features to be analyzed, and “difference” in functional features and / or structural features to be analyzed are considered. One feature is that comprehensive interaction analysis is performed by both analysis (local model). The verification result of the effectiveness of the analysis processing by the combination of the global model in the upper node and the local model in the lower node is, for example, as described in the item “2. Verification of analysis processing method”.

5). Interaction analysis processing 5-1. First Embodiment FIG. 20 is a flowchart of an interaction analysis processing program as a first embodiment executed by the CPU 10 of the interaction analysis apparatus 500. The apparatus 500 can execute (1) prediction of a protein that interacts with a compound, (2) prediction of a compound that interacts with a protein, and (3) prediction of an interaction between a compound and a protein. It is. Hereinafter, as an example, (2) prediction of a compound that interacts with a protein will be described. Other (1) prediction of the protein interacting with the compound and (3) prediction regarding the interaction between the compound and the protein can be performed by the same processing.

  The CPU 10 of the apparatus 500 receives input of data relating to the functional characteristics and / or structural characteristics of the protein to be analyzed through the operation of the keyboard / mouse 18 by the user of the apparatus (step S201 in FIG. 20). In the embodiment, as an example of input data, structural feature data obtained by digitizing an amino acid sequence is input. The CPU 10 analyzes the presence / absence of the analysis object belonging to the classification node by the evaluation function in the classification node of the hierarchy N (initial value = 1) (S203). Specifically, the CPU 10 calculates the presence / absence (Y) of belonging to the node using the input structural feature data as the explanatory variable (X) for the evaluation function of the higher class node. The evaluation function is, for example, the evaluation function (global model) of node A shown in FIGS. 18 and 19 (see symbol 69 in FIG. 19). If the analysis target does not belong to the higher class node, the CPU 10 ends the process.

  When it is analyzed that the analysis target belongs to the higher class node, the CPU 10 analyzes whether the analysis target belongs to each classification node of the hierarchy N + 1 (S205). Specifically, the CPU 10 calculates the presence / absence (Y) of belonging to the node using the input structural feature data as the explanatory variable (X) for the evaluation function of the lower classification node. The evaluation function is, for example, an evaluation function (local model) of nodes A-1, A-2,..., A-N shown in FIGS.

  The CPU 10 determines whether or not the hierarchy N + 1 is the lowest hierarchy (lowest classification node) (S207). If the CPU 10 determines that it is not the lowest hierarchy, the CPU 10 sets N to N + 1 (S209), and repeats the processing from step S205 to a classification node further lower than the classification node analyzed when the analysis target belongs. If it is determined in step S207 that it is the lowest hierarchy, the CPU 10 outputs the analysis result of the classification node to the display 20 and ends the process (S211).

  As described above, the CPU 10 applies the structural feature data to be analyzed to the global model of the upper node, and further sequentially applies it to the local model of the lower node. As a result, the CPU 10 outputs a classification node to which the analysis target belongs, that is, a protein (or protein group) having a similar structural feature and / or functional feature to the protein to be analyzed.

  In the embodiment, an example in which an analysis result of attribution to a classification node is expressed by a binary value of 0 or 1 is shown (see FIG. 19). In other embodiments, the analysis result of the attribution to the classification node may be expressed by a score value. When expressing with a score value, for example, not only the score value of the lowest classification node to which the analysis object belongs, but also all (or a part) of classification nodes from the upper classification node to the lower classification node to which the analysis object belongs. Information reflecting the score value may be output. For example, as an analysis result, an average value of score values at all belonging classification nodes can be displayed, or a value obtained by multiplying score values at all classification nodes can be displayed. The value obtained by multiplying the score values at all the classification nodes is, for example, the score value “0.8” at the higher classification node and the score value “0.7” at the lower classification node, the analysis target is “0.56” (= 0.8 × 0.7) is displayed as the score value belonging to the lower classification node. In the case of expressing with a score value, the CPU 10 can also extract and output a record (classification node or corresponding tree) having a predetermined threshold value (for example, 0.5) or more.

  FIG. 3 shows an example of the screen configuration of the analysis result output by the processing in step 211. The contents of FIG. 3 have been described in the above item “1-2. Interface”. As illustrated in FIG. 3, the analysis result of the classification node is displayed as a tree structure in addition to not only the lowest classification node to which the analysis target is predicted but also the higher classification node including the classification node. To do. Therefore, the user of the apparatus can grasp the analysis result indicating to which classification node the analysis target belongs as a position in the entire tree structure (or a part). For example, there is a difference that the lowest class node belonging to a certain branch is the third lowest from the lowest layer of the systematic classification database 900 and the lowest class node belonging to the other branch is the second lowest from the lowest. Therefore, it is possible to easily grasp the difference in the hierarchy of the plurality of classification nodes by displaying the tree structure. As another embodiment, the CPU 10 can assign the above score values to the plurality of classification nodes and display them (not shown).

5-2. Second Embodiment In the first embodiment, an example of outputting a protein (or a protein group) having similar structural characteristics and / or functional characteristics to a protein to be analyzed and an output method thereof have been described. The CPU 10 can output information on a compound that interacts with the protein (or protein group) based on information on a protein (or protein group) having similar structural characteristics and / or functional characteristics. In the following description, an example in which such interaction information is output will be described as a second embodiment.

  FIG. 21 is a flowchart of the interaction analysis processing program as the second embodiment, which is executed by the CPU 10 of the interaction analysis apparatus 500. The second embodiment and the first embodiment are common to the processing up to step S211 in FIG.

  After the process of step S211 in FIG. 20, the CPU 10 analyzes the information of the classification node of each hierarchy to which the analysis target belongs, and records it as “classification node analysis result” in the memory 12 or the like (step S301). The analysis of classification node information includes, for example, the above-described assignment of score values, or extraction of classification nodes based on threshold values. Based on the classification node analysis result, the CPU 10 records the ID of the protein belonging to the classification node corresponding to the lowest layer of the branch to which the analysis target belongs as “candidate ID” in the memory 12 or the like (S303). Specifically, the CPU 10 refers to the systematic classification database 900 illustrated in FIG. 18 and acquires “protein ID” corresponding to “node number” included in the classification node analysis result as “candidate ID”. The CPU 10 refers to each of the protein database 600, the interaction database 800, and the compound database 700, so that information on the compound that interacts with the protein specified by the “candidate ID” is stored as “interaction candidate information” in the memory 12 and the like. (S305). For example, when the candidate ID is “P001”, the CPU 10 acquires the compound “C005” that interacts with “P001” based on the interaction database 800 (see FIG. 17), and the compound database 700 (see FIG. 16). Based on the information, information on “C005” is acquired as “interaction candidate information”. CPU10 outputs interaction candidate information to the display 20, and complete | finishes a process (S307).

  22A and 22B are screen display examples output by the second embodiment. FIG. 22A is an example of a screen on which information on a protein that is predicted to interact with a compound when the information on the compound is input as an analysis target is displayed. FIG. 22B is an example of a screen on which information on a compound predicted to interact with the compound when protein information is input as an analysis target is displayed.

  In the second embodiment, the CPU 10 executes the output process (see 22 as an output example) in step S307 in FIG. 21 in addition to the output process (see FIG. 3 as an output example) in step S211 in FIG. As another embodiment, the output process in step S211 in FIG. 20 can be omitted. In addition, as the output of the analysis result according to the second embodiment, it is possible to employ display of interaction candidate information associated with the tree structure illustrated in FIG. 3 described above. Specifically, a compound (or protein) that interacts with a protein (or compound) belonging to the classification node is also displayed near the classification node in the tree structure illustrated in FIG.

6). Other Embodiments 6-1. System Configuration In the embodiment, the interaction analysis apparatus 500 is illustrated as an embodiment of the system or method of the present invention. The method of the present invention can also be used as a stand-alone normal application software. Other embodiments include the following examples.

(1) Client / Server Type As an embodiment of the system or method of the present invention, a server device that executes processing similar to that of the interaction analysis device 500, processing for transmitting data related to an analysis target, and processing for receiving analysis results ( A combination (client / server type) with a client computer that executes steps S201 and S211 in FIG. 20 may be employed. Examples of the client / server type include a system connected by a local area network (LAN) and a system using an ASP (Application Service Provider) service.

(2) Module type The system or method of the present invention can also be employed as a module for adding functions to amino acid sequence analysis software and chemical structure analysis software. In addition, the system or method of the present invention may be applied as a module for adding a function to a protein database (for example, PDB, FAMSBASE) or a chemical structural formula database (for example, ISISBase (trademark) or Accel for Excel (trademark)). it can.

  In the embodiment, the interaction analysis apparatus 500 is illustrated as an embodiment of the system or method of the present invention. As other embodiments, other devices such as a Personal Digital Assistant (PDA) may be used.

6-2. Program Execution Method In this embodiment, a program for operating the CPU 10 is stored in the hard disk 22, but this program may be read from a CD-ROM storing the program and installed in the hard disk or the like. In addition to the CD-ROM, programs such as a DVD-ROM, a flexible disk (FD), and an IC card may be installed from a computer-readable recording medium. Further, the program can be downloaded using a communication line. Also, by installing the program from the CD-ROM, the program stored in the CD-ROM is not directly executed by the computer, but the program stored in the CD-ROM is directly executed. You may do it.

  Note that programs that can be executed by a computer are not only those that can be directly executed by simply installing them, but also those that need to be converted to other forms once (for example, those that have been compressed). In addition, those that can be executed in combination with other module parts are also included.

  In each of the above embodiments, each function of FIG. 12 is realized by a CPU and a program, but a part or all of each function may be configured by hardware logic (logic circuit).

  The summary of the present invention and the preferred embodiments of the present invention have been described above. However, the terms are used for explanation rather than for limitation, and are used in the technical field related to the present invention. Those skilled in the art can recognize and implement other variations of the systems, devices, and methods within the scope of the present description. Accordingly, such variations are considered to fall within the scope of the present invention.

Target Library design for any protein (s), multi- or selective drug design for any protein (s), reverse proteomic support for any compound (s), toxicity and / or side effect prediction for any compound (s), any It is useful for predicting interactions between proteins and compounds and providing input information for various network models (diseases, side effects, toxicity, etc.). A series of operations can be automated, and as the database is expanded (commercial DB, in-house pharmacological evaluation results, reverse proteomics information, etc.), the model is updated from time to time to improve the quality of the fine copy database and the accuracy of prediction. On the other hand, by combining with information on the relationship between diseases and in vivo factors, such as DNA chip analysis, proteome research, and network models that integrate such information, we aim to develop the relationship between compounds and diseases.

Claims (14)

  Based on protein amino acid sequence information, protein amino acid sequence information systematically classified by function and / or structural similarity, compound structure information, and data that correlates protein and compound interaction information, A method for predicting the interaction between any protein and compound. In a method for predicting the interaction between any protein and any compound,
(A) a structure-activity relationship model capable of discriminating, with respect to an arbitrary compound group, a compound group that interacts with the protein and a protein group that is functionally and / or structurally similar to the protein;
(B) a structure-activity relationship model capable of discriminating a group of compounds interacting with the protein from a group of compounds interacting with the protein and a protein group that is functionally and / or structurally similar to the protein; A method of predicting by combining. In a method for predicting the interaction between an arbitrary protein and an arbitrary compound, based on systematic classification according to the function and / or structural similarity of amino acid sequence information,
(A) a structure-activity relationship model that can identify a compound group that interacts with a protein group belonging to the highest classification item to which the protein belongs to any compound group;
(B) a compound group that interacts with a protein group belonging to the classification item to which the protein belongs in each classification item lower than the highest classification item to which the protein belongs has a common parent classification item with the protein A method of predicting by combining with a structure activity model that can be identified for a compound group that interacts with a protein group belonging to a child classification item.   A computer-readable program for causing a computer to execute the method according to claim 1.    Based on protein amino acid sequence information, protein amino acid sequence information systematically classified by function and / or structural similarity, compound structure information, and data that correlates protein and compound interaction information, A system that predicts the interaction between any protein and compound. In a system that predicts the interaction between any protein and any compound,
(A) a structure-activity relationship model capable of discriminating, with respect to an arbitrary compound group, a compound group that interacts with the protein and a protein group that is functionally and / or structurally similar to the protein;
(B) a structure-activity relationship model capable of discriminating a group of compounds interacting with the protein from a group of compounds interacting with the protein and a protein group that is functionally and / or structurally similar to the protein; System that predicts by combining. In a system for predicting the interaction between any protein and any compound, based on systematic classification by amino acid sequence information function and / or structural similarity,
(A) a structure-activity relationship model that can identify a compound group that interacts with a protein group belonging to the highest classification item to which the protein belongs to any compound group;
(B) a compound group that interacts with a protein group belonging to the classification item to which the protein belongs in each classification item lower than the highest classification item to which the protein belongs has a common parent classification item with the protein A system that predicts in combination with a structure activity model that can identify a group of compounds that interact with a group of proteins belonging to a child category. A prediction system for predicting proteins of similar function and / or structure,
The prediction system is
(A) a first record for recording information on a first class protein belonging to the first class indicating a classification of functional and / or structural features of the protein and information on a non-first class protein not belonging to the first class means,
(B) a second class belonging to a second class indicating a class of functional features and / or structural features which are selected from the first class proteins recorded in the first recording means and which are sub-concepts of the first class. Second recording means for recording information on two-class protein and non-second-class protein belonging to the first class but not belonging to the second class;
(C) acquisition means for acquiring prediction target information indicating the functional characteristics and / or structural characteristics of the prediction target protein;
(D) Based on the prediction target information acquired by the acquisition means and the information recorded in the first recording means, the prediction target protein and the first classification protein in the comparison with the non-first classification protein A first analysis means for analyzing the similarity of
(E) When it is analyzed by the first analysis means that the protein to be predicted is similar to the first classification protein, further, based on the information to be predicted and information recorded in the second recording means A second analysis means for analyzing the similarity between the protein to be predicted and the second classification protein in comparison with the non-second classification protein;
(F) an output means for outputting information on a protein having a functional feature and / or a structural feature similar to those of the prediction target protein, based on the analysis results by the first analysis means and the second analysis means;
Prediction system with The (a) protein information recorded in the first recording means and / or (b) the protein information recorded in the second recording means of the prediction system is associated with the information of the compound that interacts with the protein. And
The prediction system further includes:
(G) Based on information on a compound that interacts with the protein and information on a protein similar to the protein to be predicted analyzed by (d) the first analysis means and / or (e) the second analysis means. , An interaction information analysis means for analyzing information of a compound predicted to interact with the protein to be predicted,
With
The (f) output means includes:
Outputting information on a protein having functional characteristics and / or structural characteristics similar to the protein to be predicted, and / or information on a compound analyzed by the interaction information analysis means,
The prediction system of claim 8. The (f) output means includes:
In addition to the information on the protein and / or the compound that interacts with the protein as the analysis result by the second analysis means, the concept as the analysis result by the first analysis means, which is a larger concept than the second classification. Outputs information on protein and / or compound that interacts with the protein.
The prediction system according to claim 8 or 9. A prediction system for predicting proteins of similar function and / or structure,
The prediction system is
(A) a first record for recording information on a first class protein belonging to the first class indicating a classification of functional and / or structural features of the protein and information on a non-first class protein not belonging to the first class apparatus,
(B) a second class belonging to a second class indicating a class of functional features and / or structural features selected from the first class proteins recorded in the first storage device and having a concept smaller than the first class. A second recording device for recording information on two-class protein and non-second-class protein belonging to the first class but not the second class;
With
The central processing unit (CPU) of the prediction system is
(C) obtaining prediction target information indicating the functional characteristics and / or structural characteristics of the prediction target protein;
(D) Similarity between the prediction target protein and the first classification protein in comparison with the non-first classification protein based on the acquired prediction target information and the information recorded in the first recording device Analyze
(E) When it is analyzed that the protein to be predicted is similar to the protein of the first classification, the non-second classification is further performed based on the information to be predicted and information recorded in the second recording device. Analyzing the similarity between the protein to be predicted and the second class protein in comparison with the protein;
(F) based on the analysis result, outputting information about a protein having a functional characteristic and / or a structural characteristic similar to the prediction target protein to a display device;
A prediction system characterized by A computer-readable program for causing a computer to function as a prediction system for predicting a protein having a similar function and / or structure,
The program causes the computer to
(A) a first record for recording information on a first class protein belonging to the first class indicating a classification of functional and / or structural features of the protein and information on a non-first class protein not belonging to the first class means,
(B) a second class belonging to a second class indicating a class of functional features and / or structural features which are selected from the first class proteins recorded in the first recording means and which are sub-concepts of the first class. Second recording means for recording information on two-class protein and non-second-class protein belonging to the first class but not belonging to the second class;
(C) acquisition means for acquiring prediction target information indicating the functional characteristics and / or structural characteristics of the prediction target protein;
(D) Based on the prediction target information acquired by the acquisition means and the information recorded in the first recording means, the prediction target protein and the first classification protein in the comparison with the non-first classification protein A first analysis means for analyzing the similarity of
(E) When it is analyzed by the first analysis means that the protein to be predicted is similar to the first classification protein, further, based on the information to be predicted and information recorded in the second recording means A second analysis means for analyzing the similarity between the protein to be predicted and the second classification protein in comparison with the non-second classification protein;
(F) an output means for outputting information on a protein having a functional feature and / or a structural feature similar to those of the prediction target protein, based on the analysis results by the first analysis means and the second analysis means;
Program to function as a prediction system with A prediction system that predicts proteins (or compounds) that are similar in function and / or structure,
The prediction system is
(A) Information on the first class protein (or first class compound) belonging to the first class indicating the classification of the functional characteristics and / or structural characteristics of the protein (or compound), and the non-first that does not belong to the first class First recording means for recording information on classified proteins (or non-first classified compounds);
(B) a functional feature and / or structural feature classification selected from the first classification proteins (or first classification compounds) recorded in the first recording means and having a concept smaller than the first classification; Information of a second class protein (or second class compound) belonging to the second class shown, and a non-second class protein (or non-second class compound) belonging to the first class but not belonging to the second class Second recording means for recording,
(C) an acquisition means for acquiring prediction target information indicating a functional characteristic and / or a structural characteristic of a prediction target protein (or prediction target compound);
(D) The prediction target in comparison with the non-first classification protein (or non-first classification compound) based on the prediction target information acquired by the acquisition means and the information recorded in the first recording means First analysis means for analyzing the similarity between a protein (or a prediction target compound) and the first class protein (or first class compound);
(E) When the first analysis means analyzes that the prediction target protein (or prediction target compound) is similar to the first classification protein (or first classification compound), the prediction target information and the Based on the information recorded in the second recording means, in the comparison with the non-second classification protein (or non-second classification compound), the prediction target protein (or prediction target compound) and the second classification protein ( Or a second analysis means for analyzing the similarity with the second classification compound),
(F) Based on the analysis results by the first analysis means and the second analysis means, information on a protein (or compound) having a functional characteristic and / or a structural characteristic similar to the prediction target protein (or prediction target compound) is output. Output means,
Prediction system with A prediction method for predicting proteins having similar functions and / or structures,
The prediction method is:
(A) Information on the first class protein belonging to the first class indicating the classification of the functional characteristics and / or structure characteristics of the protein and information on the non-first class protein not belonging to the first class are used as the first recording means. Record,
(B) a second class belonging to a second class indicating a class of functional features and / or structural features which are selected from the first class proteins recorded in the first recording means and which are sub-concepts of the first class. Record the information on the second class protein and the non-second class protein belonging to the first class but not belonging to the second class in the second recording means;
(C) obtaining prediction target information indicating the functional characteristics and / or structural characteristics of the prediction target protein;
(D) Based on the prediction target information acquired by the acquisition means and the information recorded in the first recording means, the prediction target protein and the first classification protein in the comparison with the non-first classification protein Analyze the similarity of
(E) When it is analyzed that the prediction target protein is similar to the first classification protein by analyzing the similarity between the prediction target protein and the first classification protein, the prediction target information and the second classification protein Based on the information recorded in the recording means, the similarity between the protein to be predicted and the second classification protein in the comparison with the non-second classification protein is analyzed,
(F) Based on the analysis result, information on a protein having functional characteristics and / or structural characteristics similar to those of the prediction target protein is output.
Prediction method.

Images